MATLAB Environmentbayanbox.ir/.../06-Neural-Networks-Using-Matlab.pdfVariables MATLAB is a weakly...

transcript

MATLAB Environment

Neural Networks - Shahrood University - Hossein Khosravi

MATLAB Help

help sin

Inline help

Concise

doc sin

Opens help browser

Comprehensive

Several Examples

Writing scripts using editor

Remarks

COMMENT! Anything following a % is seen as a comment

The first contiguous comment becomes the script's help file

Comment thoroughly to avoid wasting time later

Note that scripts are somewhat static, since there is no input and no explicit output

All variables created and modified in a script exist in the workspace even after it has stopped running

Variables

MATLAB is a weakly typed language No need to initialize variables!

MATLAB supports various types; The most often used are » pi_num = 3.14

64-bit double (default)

» a_char = ‘a’ 16-bit char

Most variables you’ll deal with will be vectors or matrices of doubles or chars

Other types are also supported: complex, symbolic, 16-bit and 8 bit integers, etc.

Matrices: The most common type

Built in functions

MATLAB has an enormous library of built-in functions

Almost any function you think, is available.

Call using parentheses –passing parameter to function:

»sqrt(2)

»log(2), log10(0.23)

»cos(1.2), atan(-.8)

»exp(2+4*i)

»round(1.4), floor(3.3), ceil(4.23)

»angle(1+2i); abs(1+2i);

Example: Activation functions used in NN's

% Illustration of various activation functions used in

x = -10:0.1:10;

tmp = exp(-x);

y1 = 1./(1+tmp);

y2 = (1-tmp)./(1+tmp);

y3 = x;

subplot(131); plot(x, y1); grid on;

title('Logistic Function');

title('Hyperbolic Tangent Function');

title('Identity Function');

Output

-10 -5 0 5 10-2

2Logistic Function

-10 -5 0 5 10-2

2Hyperbolic Tangent Function

-10 -5 0 5 10-10

10Identity Function

Example: Curve fitting

% Polynomial fit

x=-4:0.1:4;

y=x.^2;

y=y+randn(size(y));

plot(x,y,'.');

p = polyfit(x,y,2)

hold on;

plot(x,polyval(p,x),'r');

-4 -3 -2 -1 0 1 2 3 4-5

User functions

No need for return:

MATLAB 'returns' the variables whose names match those in the function declaration

Variable scope:

Any variables created within the function but not returned disappear after the function stops running function

Relational Operators

MATLAB uses mostly standard relational operators Equal ==

Not equal ~=

greater than >

less than <

greater or equal >=

less or equal <=

Logical operators element-wise scalars And & &&

Or | ||

Xor xor

All true all

Any true any

Boolean values: zero is false, nonzero is true

See help . for a detailed list of operators

Code Efficiently

Avoiding Loops

Neural Networks

nnstart

nntool

nftool

nprtool

nctool

ntstool

newp, newhop, newff, …

Neural Nets Using MATLAB

nnstart: A good point to start with neural network toolbox

Example: fitnet

%fit net

%[x,t] = simplefit_dataset;

[x t] = simplefit_create;

%subplot(211)

plot(x,t,'.');

net = fitnet(5); % I told everything you think may be foundnet.trainParam.epochs = 5;

net = train(net,x,t);

%view(net);

y = net(x);

%subplot(212)

hold on

plot(x,y,'r');

Result

0 1 2 3 4 5 6 7 8 9 100

5 Epochs 10 Epochs

Previous Example using Fitting Tool (nftool)

Here we can divide dataset into

train, test and validation

Results

Pattern Recognition Tool (nprtool)

MSE Graph

Perceptron

newp: Create a perceptron.

Obsoleted in R2010b NNET 7.0.

Last used in R2010a NNET 6.0.4.

Run digit recognition program: newp_digits.m

Adaptive Linear Network: Adaline

newlin

Or simple coding!

Example: Prediction

Run program Adalline_Prediction.m

Example: System Identification

Run program Adaline_System_Identification.m

Example: Digit recognition

Run program ReadFeatures.m and train.m

MATLAB Environmentbayanbox.ir/.../06-Neural-Networks-Using-Matlab.pdfVariables MATLAB is a weakly...

Documents